AI Governance
AI Governance in ANTS Platform shows you which AI agents you run, what they touch, what risky content passes through them, and how far each one is from meeting your regulatory frameworks. Everything is built from the traces your agents send, plus the governance evidence your team records.
AI Governance features need the Enterprise governance entitlement (Enterprise plan). Without it, the pages show a preview. What each person can see and do depends on their role. See Members & Roles.
Where to find it
| Sidebar group | Pages |
|---|---|
| Discover AI | AI Agent Center, AI TrailMaps |
| Gauge AI | AI Inventory |
| Governance | Risk Dashboard, AI Guardrails, AI Violations, My Requests, Available Requests, Evidence Collection, Framework Coverage, Audit & Export |
Capabilities
| Capability | What it does |
|---|---|
| AI Agent Center | A catalog of every agent with requests, cost, latency, errors, risk score, compliance violations and detections, and one-click notifications to your integrations |
| AI TrailMaps | A map of the models, tools, MCP servers, datastores and external services each agent reached, and where data left the organization |
| AI Inventory | The agents, models, frameworks and AI tools in use, with model approval |
| AI Guardrails | Per-agent policies that check prompts and responses inline (log, warn, block or redact, chosen per detection category) through the Python SDK, the REST API, the Portkey connector or the Budget AI gateway, with 80 pre-built detectors including secrets and credentials, source code and validated card numbers and SSNs, plus custom regular expression and keyword detectors |
| PII Detection | Scheduled scans of stored traces for PII, PCI, PHI, secrets and credentials, source code, harmful content, bias, jailbreak attempts and custom detector matches, with PagerDuty alerting |
| AI Violations | Every detection, linked to the trace it came from, with optional false-positive review and a per-detector accuracy report |
| Risk Dashboard | Risk scores from detection, unapproved-model and error rates, a likelihood by impact heatmap and four risk dimensions |
| Compliance | Governance profiles, evidence in eight categories with attestations, review and approval through Slack, Jira or Linear, and pre-built control mappings for the EU AI Act, ISO/IEC 42001, NIST AI RMF, SOC 2, ISO/IEC 27001 and NIST CSF 2.0 |
| Drift Alerts | Gaps between an agent's declared evidence and what its traces show, recalculated on a background schedule and sent automatically to your integrations |
| Audit & Export | PDF and Excel compliance reports and the governance activity history |
| Notifications | Compliance, detection and drift notifications to Slack, Jira, Linear, ServiceNow and PagerDuty |
Get started
Instrument your agents
Send traces from your agents with the ANTS Platform SDKs or OpenTelemetry, and give each agent a name. Agents then appear in AI Agent Center automatically. Trace connectors can also bring in agents from cloud platforms. See AI Inventory.
Turn on detection
In Governance → AI Guardrails, create a Scheduled Detection policy for each agent so its traces are scanned. See PII Detection. For inline blocking or redaction, use Real-time Enforcement. See AI Guardrails.
Approve your models
Approve the models your organization allows in the AI Inventory. Use of any other model is counted as a compliance violation.
Connect your integrations
Connect Slack, Jira, Linear, ServiceNow and PagerDuty, and map agents to channels, projects and teams, so requests and alerts reach the right people.
Build governance profiles
Create a governance profile per agent in Evidence Collection, collect evidence, and approve it. See Compliance.
Enforce guardrails in code
Guardrail policies are configured in Governance → AI Guardrails. Your application enforces them at runtime with the ants-platform Python SDK. Pass the project keys as one publicKey:secretKey string and the agent's ID from the policy:
Each check returns PASS, WARNING, BLOCKED or SANITIZED. Drop-in wrappers for OpenAI, Anthropic, AWS Bedrock, Google Gen AI and Vertex AI, and the REST endpoint for other languages, are covered in AI Guardrails.
Policies, governance profiles, evidence and reports are managed in the product, not through the SDK.
Who can do what
| Role | AI Governance access |
|---|---|
| Owner | Everything |
| Security & Compliance | Administers Governance: guardrail policies, evidence, requests and approvals, drift alerts, AI Inventory, report export. Reads AI Agent Center and AI TrailMaps. |
| Department Admin | Reads Governance pages. Uses AI Agent Center and AI TrailMaps. No Audit & Export. |
| Auditor | Reads Governance pages and AI TrailMaps. Views the audit trail and downloads existing reports. |
| Cost Analyst | Uses AI Agent Center. No Governance pages. |
| Manager Viewer, End User, None | No AI Governance access |
Organizations can also define custom roles. See Custom Roles.